Statistical Background Subtraction for a Mobile Observer

Statistical Background Subtraction for a Mobile Observer,10.1109/ICCV.2003.1238315,Eric Hayman,Jan-olof Eklundh

Statistical Background Subtraction for a Mobile Observer   (Citations: 48)
BibTex | RIS | RefWorks Download
Statistical background modelling and subtraction has proved to be a popular and effective class of algorithms for segmenting independently moving foreground objects out from a static background, without requiring any a priori in- formation of the properties of foreground objects. This pa- per presents two contributions on this topic, aimed towards robotics where an active head is mounted on a mobile ve- hicle. In periods when the vehicle's wheels are not driven, camera translation is virtually zero, and background sub- traction techniques are applicable. Parts of this work are also highly relevant to surveillance and video conferencing. The first part of the paper presents an efficient proba- bilistic framework for when the camera pans and tilts. A unified approach is developed for handling various sources of error, including motion blur, sub-pixel camera motion, mixed pixels at object boundaries, and also uncertainty in background stabilisation caused by noise, unmodelled ra- dial distortion and small translations of the camera. The second contribution regards a Bayesian approach to specifically incorporate uncertainty concerning whether the background has yet been uncovered by moving foreground objects. This is an important requirement during initiali- sation of a system. We cannot assume that a background model is available in advance since that would involve stor- ing models for each possible position, in every room, of the robot's operating environment. Instead the background model must be generated online, very possibly in the pres- ence of moving objects.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
Sort by: